Bedside
Bedside is an AI system that automates routine hospital-to-family communication while enforcing a hard boundary on what AI is allowed to deliver.
Core trick
Classifies each chart-derived fact into safe, context-needed, or acute tiers, auto-texting only safe facts and locking the channel for acute events.
Description
Bedside is an AI system that automates routine hospital-to-family communication while enforcing a hard boundary on what AI is allowed to deliver. The problem: nurses spend 30+ minutes a day fielding repetitive phone calls from patients' families asking for status updates, and families of hospitalized patients have no reliable way to get information without calling the unit. Bedside takes a short nurse dictation plus the patient's FHIR chart data and runs every fact through an AI classifier with three outcomes: safe facts are automatically texted to family members in their own language, facts requiring context trigger only a nurse-approved "a nurse will call you" notice, and serious findings lock the texting channel entirely, the system schedules a human phone call and gives the nurse a briefing of what the family already knows and has asked. Family members can text questions back: the system answers only from already-shared facts or verifiable events (like whether a doctor has visited), and routes all clinical questions to the care team using a fixed, unvarying reply. The result is that nurses stop making routine update calls, families get proactive information instead of chasing it, and bad news is never delivered by AI, only by a human voice.
README excerpt
# Bedside Built solo at The Future of Agentic AI in Healthcare hackathon (Abridge x Anthropic x Lightspeed), July 18, 2026. An agentic judgment layer that turns inpatient clinical data into proactive family updates — and knows when to say nothing. Green-tier facts become plain-language texts to the care circle. Actionable items promise a human call. Acute events lock the channel: bad news only ever travels by human voice, and the agent's job is to make sure the nurse walks into that call prepared.  ## Provenance - `synthetic-ambient-fhir-25.json` — synthetic dataset provided by Abridge for hackathon use. Not my work. - Everything else in this repo — the judgment/verification pipeline, tier logic, channel lock, and both interfaces — was designed and built during the event, today. See commit history. - Demo presents two encounter days from the synthetic record with illustrative day labels, ordered stable-then-deterioration to reflect a typical clinical course; underlying data dates are in the code. ## What was built today (all of it) - FHIR data loader over the Abridge synthetic record (patient 7bd9e5b0, COVID inpatient admission) - Two-pass AI judgment layer: tier classification (green/yellow/red) with per-fact reasons, asymmetric uncertainty rule (uncertain → more restrictive tier) - Tier-based oversight: green auto-sends, yellow requires nurse approval of a fixed callback template, red locks the family channel - Per-recipient gen